Re: OpenBSD frexpl failure

[Bruno Haible]

Eric Blake wrote:
> I'm wondering if we should also add a configure-time compile test in
> float_h.m4, which does the sanity check of this untested snippet:
> 
> AC_CHECK_SIZEOF([double])
> AC_CHECK_SIZEOF([long double])
> AC_COMPUTE_INT([variable],
>   [(SIZEOF_DOUBLE == SIZEOF_LONG_DOUBLE)
>    == (DBL_MANT_DIG == LDBL_MANT_DIG)], [<float.h>])

This test is not right. You can have
   SIZEOF_DOUBLE < SIZEOF_LONG_DOUBLE && DBL_MANT_DIG == LDBL_MANT_DIG
(the FreeBSD 6.1 case that we had recently). You can also have
   SIZEOF_DOUBLE == SIZEOF_LONG_DOUBLE && DBL_MANT_DIG < LDBL_MANT_DIG
(if a platform's minimum alignment for types is 16 bytes).

You could try to compute LDBL_MANT_DIG through constant expressions in the
compiler, but this is more likely to expose compiler bugs (especially when
cross-compiling) than to help.

Bruno